SOFIE - A Unified Approach To Ontology-Based Information Extraction Using Reasonig

SOFIE - A Unified Approach To Ontology-Based Information Extraction Using ReasonigTobias WunnerUnit for Natural Language Processing (UNLP)firstname.lastname@deri.orgWednesday,22nd June, 2011DERI, Reading Group1

Based On:“SOFIE: A Self-Organizing Framework for Information Extraction”Authors: Fabian Suchanek, Mauro Sozio, Gerhard WeikumPublished: World Wide Web Conference (WWW) Madrid, 20092

OverviewIntroductionSOFIE Model + RulesExcursion: SatisfiabilitySOFIE ApproachEvaluation experimentsConclusion3

MotivationClassical IE on text pattern-based  80pcSemistructural approach Wikipedia infoboxes 95%Idea of Paper: combine use text (hypotheses) + ontology (trusted facts)4

Example5Document1YAGO ontologyfamilyName(AlbertEinstein, Einstein)bornIn(AlbertEinstein, Germany)attendedSchoolIn( AlbertEinstein, Germany)Einstein attended secondary school in Germany.New Knowledge

General IdeaExpress extraction patterns as factRules to understand usage of termsAdd restrictions6patternOcc(“X went to school in Y”,Einstein, Switzerland)patternOcc(Pattern,X,Y) and R(X,Y) ⇒ express(Pattern,R)

ContributionUnified approach toPattern matchingWord Sense DisambiguationReasoningLarge ScaleOn Unstructured Data7

Pattern extraction with WICsExtract patterns based on ‘interesting’ entities8DocumentsEinstein was born at Ulm in Württemberg, Germany, on March 18, 1879. When Albert was around four, his father gave him a magnetic compass. When Albert became older, he went to a school in Switzerland. After he graduated, he got a job in the patent office there…Knowledge BasepatternOcc(“Einstein was born in Ulm”,Einstein@D1, Ulm@D1) [1]patternOcc(“Ulm is in Württemberg, Germany”,Ulm@D1, Germany@D1) [1]patternOcc(“Albert .. Switzerland”,Albert@D1, Switzerland@D1) [1]WICs (Word in Context)

GroundingTest RulesHow?find an instance which satisfies the formulae9bornIn(Einstein,Ulm) ⇒ ¬bornIn(Einstein,Timbuktu)studiedIn(Einstein,Ulm)bornIn(X,Ulm) ⇒ ¬bornIn(X,Timbuktu)studiedIn(X,Ulm)

Rules (Hypotheses)DisambiguationdisambiguatesAs(Albert@D,AlberEinstein)[?]Expresses a new factexpresses(P, livedIn(Einstein,Switzerland) )[?]New factsCityIn(Ulm,Germany)[?]10

New fact rule...with disambiguation11“Pattern P expresses Relation R when analysis of WICs are disambiguated”patternOcc( P, WX, WY ) anddisambiguatesAs(WX, X) anddisambiguatesAs(WY, Y) andR(X,Y)⇒ express( P, R )

RestrictionsDisambiguation disambiguation prior should influence choice of disambiguation12N - any disamb. functiondisambPrior( W, X, N )⇒ disambiguatedAs( W, X )| words(D1) ∩ rel(AlbertEinstein)|| words(D1) |

RestrictionsFunctional restrictions13R(X,Y) and type(R, function) anddifferent(Y,Z)⇒ ¬R(X,Z)“Albert@D1 born in?”Albert@D1 ≠ Albert@D2

SOFIE RulesFramework to test the hypothesesQuestion “How to satisfy all them?” rules + trusted facts14dismbPrior(Albert@D1, AlbertEinstein, 10)⇒ disambiguatesAs(Albert@D1, AlbertEinstein)patternOcc( P, X, Y ) andR(X,Y)⇒ express( P, R )dismbPrior(Albert@D1, HermannEinstein, 3)⇒ disambiguatesAs(Albert@D1, HermannEinstein) Country(Germany)livedIn(AlbertEinstein,Ulm) …

SAT / MAX SATSAT (Satisfiability)proove formula can be TRUEComplexity ClassesP  Good example: NkNP  Bad cNe.g. naive algorithm for 100 variables 2100 x 10-10 ms per row = 4 x 1012 yNot always.. 3SAT in (4/3)NSAT Solver15F = (X or Y or Z) and (¬X or Y or Z) and (¬X or ¬Y or ¬Z)G = (X or Y) and (¬X or ¬Y) and (X)truth table has 23 rowsDetails Schöning 2010

SAT / MAX SATSAT (Satisfiability)proove formula can be TRUEComplexity ClassesP  Good example: NkNP  Bad cNe.g. naive algorithm for 100 variables 2100 x 10-10 ms per row = 4 x 1012 yNot always.. 3SAT in (4/3)NSAT SolverMAX SAT16F = (X or Y or Z) and (¬X or Y or Z) and (¬X or ¬Y or ¬Z)G = (X or Y) and (¬X or ¬Y) and (X)truth table has 23 rowsDetails Schöning 2010

Weighted MAX SAT in SOFIE...back to SOFIEthis is MAX SAT but with weights17rules + trusted facts Country(Germany)livedIn(AlbertEinstein,Ulm) …dismbPrior(Albert@D1, AlbertEinstein, 10)⇒ disambiguatesAs(Albert@D1, AlbertEinstein)patternOcc( P, X, Y ) andR(X,Y)⇒ express( P, R )dismbPrior(Albert@D1, HermannEinstein, 3)⇒ disambiguatesAs(Albert@D1, HermannEinstein)

Weighted MAX SAT in SOFIEWeighted MAX SAT is NP hardonly approximation algorithms impractical to find optimal solutionSAT SolverJohnson’s algorithm:  2/3 (apprx guarantee)

Weighted MAX SAT in SOFIEFunctional MAX SATSpecialized reasoning (support for functional properties)Approximation guarantee 1/2Propagates dominating unit clausesConsiders only unit clausesA v B [w1]A v B [w2]B v C [w3]C [w4]A v B [10]A [10]A [30]A = true30 > 10+10

Controlled experimentCorpus from Wikipedia infoboxes100 articlesSemantic is known!20

Controlled experimentLarge-scale: Corpus from Wikipedia articles2000 articles13 frequent relations from YAGOParsing = 87min Reaoning = 77min21

Unstructured text sources150 news paper articlesrelation under test headquarterOfYAGO (modified with relation seeds)Parsing 87min WeightedMaxSat 77mindisambiguated entries (provenance) could be manually assessed22functionalrelation

Unstructured text sourcesLarge-scale:10 biographies for each of 400 US senators5 relationshipsDisambiguation was not ideal for YAGO (13 James Watson)Parsing 7h W-MAX-SAT 9hResults4 good1 bad (misleading patterns)23

MAX SAT can’t do OWL per se (Open World Assumption)Reformulate OWL in propositional logicOWL  FOL  Skolem Normal Form  Propositional LogicMight find OWL-inconsistent ontologies due to OW Assumption24define a student as a subclass “attends some course”⇒ ∀ x, ∃ y: attends(x,y), Course(y) -> Student(y)⇒ ∀ x: attends(x,k), Course(y) -> Student(y); ∃ k⇒ ¬attends(xi, ki) or ¬Course(xi) or Student(xi); k=x1 .. xnInferred Ontology{ Student(alex), Student(bob), Student subClassOf attends some Course, attends(alex, SemanticWeb) }Details JMC 2010

ConclusionsOntology-based IE (OBIE) reformulated as weighted MAX SAT problemApproximation algorithm with 1/2Works and scales (large corpus + YAGO)25

LimitationsSpecialized approximation algorithmAccounts for SOFIE rules NOT OWLMAX SAT Restrictions∈ Prepositional Logic∉ First-Order LogicOntology population approach (can’t infer new relations)26

References27F Suchanek et al, SOFIE: a self-organizing framework for information extraction, Proceeding WWW '09 Proceedings of the 18th international conference on World wide web, linkJohn McCrae, Automatic Extraction Of Logically Consistent Ontologies From Text, PhD thesis at National Institute of Informatics, Japan, 2009 linkUwe Schöning: Das SAT-Problem. In Informatik Spektrum 33(5): 479-483, 2010, linkF Suchanek, Automated Construction and Growth of a Large Ontology, PhD thesis at Technology of Saarland University. Saarbrücken, Germany, 2009, link

SOFIE - A Unified Approach To Ontology-Based Information Extraction Using Reasonig

More Related Content

What's hot (20)

Similar to SOFIE - A Unified Approach To Ontology-Based Information Extraction Using Reasonig (20)

Recently uploaded (20)

SOFIE - A Unified Approach To Ontology-Based Information Extraction Using Reasonig